Reduced-order models (ROMs) have become a key technology for the efficient simulation of complex systems governed by partial differential equations. By constructing low-dimensional representations of high-fidelity models, ROMs enable rapid evaluations in many-query settings such as optimization, control, and uncertainty quantification, and are central to the development of digital twins. However, ensuring accuracy and robustness in strongly nonlinear and multi-parameter regimes remains a major challenge for classical approaches based on linear subspaces. Recent advances in scientific machine learning, particularly generative methods but also broader learning-based approaches, open new perspectives for reduced-order modeling. These approaches provide flexible tools to learn complex solution distributions and nonlinear manifolds, going beyond the limitations of traditional projection-based techniques. Machine Learning techniques, such as generative models, operator learning, data-driven closure modeling, can enhance ROMs by improving the representation of nonlinear dynamics, enabling data augmentation, and supporting the development of hybrid physics–data-driven surrogates. This mini-symposium aims to bring together researchers working on reduced-order modeling, Machine Learning for physical systems, and digital twin technologies. The focus is on emerging methodologies that leverage a wide range of AI techniques, with a focus on generative models, to build accurate, robust, and scalable ROMs. Topics of interest include nonlinear representation learning, generative modeling of solution manifolds, hybrid physics–machine learning methods, uncertainty quantification, learning-based closure strategies, operator learning, and applications to fluid dynamics, structural mechanics, and multi-physics systems. By promoting contributions ranging from theoretical advances to industrial applications, this mini-symposium seeks to foster cross-disciplinary exchange and to advance the development of next-generation reduced-order modeling techniques driven by artificial intelligence and scientific machine learning.